2.11BSD#431 - /usr/src/sys/sys/uipc_socket2.c

   1: /*
   2:  * Copyright (c) 1982, 1986 Regents of the University of California.
   3:  * All rights reserved.
   4:  *
   5:  * Redistribution and use in source and binary forms are permitted
   6:  * provided that this notice is preserved and that due credit is given
   7:  * to the University of California at Berkeley. The name of the University
   8:  * may not be used to endorse or promote products derived from this
   9:  * software without specific prior written permission. This software
  10:  * is provided ``as is'' without express or implied warranty.
  11:  *
  12:  *	@(#)uipc_socket2.c	7.3 (Berkeley) 1/28/88
  13:  */
  14: 
  15: #include "param.h"
  16: #include "systm.h"
  17: #include "user.h"
  18: #include "proc.h"
  19: #include "file.h"
  20: #include "inode.h"
  21: #include "buf.h"
  22: #include "mbuf.h"
  23: #include "protosw.h"
  24: #include "socket.h"
  25: #include "socketvar.h"
  26: 
  27: /*
  28:  * Primitive routines for operating on sockets and socket buffers
  29:  */
  30: 
  31: /*
  32:  * Procedures to manipulate state flags of socket
  33:  * and do appropriate wakeups.  Normal sequence from the
  34:  * active (originating) side is that soisconnecting() is
  35:  * called during processing of connect() call,
  36:  * resulting in an eventual call to soisconnected() if/when the
  37:  * connection is established.  When the connection is torn down
  38:  * soisdisconnecting() is called during processing of disconnect() call,
  39:  * and soisdisconnected() is called when the connection to the peer
  40:  * is totally severed.  The semantics of these routines are such that
  41:  * connectionless protocols can call soisconnected() and soisdisconnected()
  42:  * only, bypassing the in-progress calls when setting up a ``connection''
  43:  * takes no time.
  44:  *
  45:  * From the passive side, a socket is created with
  46:  * two queues of sockets: so_q0 for connections in progress
  47:  * and so_q for connections already made and awaiting user acceptance.
  48:  * As a protocol is preparing incoming connections, it creates a socket
  49:  * structure queued on so_q0 by calling sonewconn().  When the connection
  50:  * is established, soisconnected() is called, and transfers the
  51:  * socket structure to so_q, making it available to accept().
  52:  *
  53:  * If a socket is closed with sockets on either
  54:  * so_q0 or so_q, these sockets are dropped.
  55:  *
  56:  * If higher level protocols are implemented in
  57:  * the kernel, the wakeups done here will sometimes
  58:  * cause software-interrupt process scheduling.
  59:  */
  60: 
  61: soisconnecting(so)
  62:     register struct socket *so;
  63: {
  64: 
  65:     so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
  66:     so->so_state |= SS_ISCONNECTING;
  67:     WAKEUP((caddr_t)&so->so_timeo);
  68: }
  69: 
  70: soisconnected(so)
  71:     register struct socket *so;
  72: {
  73:     register struct socket *head = so->so_head;
  74: 
  75:     if (head) {
  76:         if (soqremque(so, 0) == 0)
  77:             panic("soisconnected");
  78:         soqinsque(head, so, 1);
  79:         sorwakeup(head);
  80:         WAKEUP((caddr_t)&head->so_timeo);
  81:     }
  82:     so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING);
  83:     so->so_state |= SS_ISCONNECTED;
  84:     WAKEUP((caddr_t)&so->so_timeo);
  85:     sorwakeup(so);
  86:     sowwakeup(so);
  87: }
  88: 
  89: soisdisconnecting(so)
  90:     register struct socket *so;
  91: {
  92: 
  93:     so->so_state &= ~SS_ISCONNECTING;
  94:     so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
  95:     WAKEUP((caddr_t)&so->so_timeo);
  96:     sowwakeup(so);
  97:     sorwakeup(so);
  98: }
  99: 
 100: soisdisconnected(so)
 101:     register struct socket *so;
 102: {
 103: 
 104:     so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
 105:     so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
 106:     WAKEUP((caddr_t)&so->so_timeo);
 107:     sowwakeup(so);
 108:     sorwakeup(so);
 109: }
 110: 
 111: /*
 112:  * When an attempt at a new connection is noted on a socket
 113:  * which accepts connections, sonewconn is called.  If the
 114:  * connection is possible (subject to space constraints, etc.)
 115:  * then we allocate a new structure, properly linked into the
 116:  * data structure of the original socket, and return this.
 117:  */
 118: struct socket *
 119: sonewconn(head)
 120:     register struct socket *head;
 121: {
 122:     register struct socket *so;
 123:     register struct mbuf *m;
 124: 
 125:     if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2)
 126:         goto bad;
 127:     m = m_getclr(M_DONTWAIT, MT_SOCKET);
 128:     if (m == NULL)
 129:         goto bad;
 130:     so = mtod(m, struct socket *);
 131:     so->so_type = head->so_type;
 132:     so->so_options = head->so_options &~ SO_ACCEPTCONN;
 133:     so->so_linger = head->so_linger;
 134:     so->so_state = head->so_state | SS_NOFDREF;
 135:     so->so_proto = head->so_proto;
 136:     so->so_timeo = head->so_timeo;
 137:     so->so_pgrp = head->so_pgrp;
 138:     soqinsque(head, so, 0);
 139:     if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH,
 140:         (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0)) {
 141:         (void) soqremque(so, 0);
 142:         (void) m_free(m);
 143:         goto bad;
 144:     }
 145:     return (so);
 146: bad:
 147:     return ((struct socket *)0);
 148: }
 149: 
 150: soqinsque(head, so, q)
 151:     register struct socket *head, *so;
 152:     int q;
 153: {
 154: 
 155:     so->so_head = head;
 156:     if (q == 0) {
 157:         head->so_q0len++;
 158:         so->so_q0 = head->so_q0;
 159:         head->so_q0 = so;
 160:     } else {
 161:         head->so_qlen++;
 162:         so->so_q = head->so_q;
 163:         head->so_q = so;
 164:     }
 165: }
 166: 
 167: soqremque(so, q)
 168:     register struct socket *so;
 169:     int q;
 170: {
 171:     register struct socket *head, *prev, *next;
 172: 
 173:     head = so->so_head;
 174:     prev = head;
 175:     for (;;) {
 176:         next = q ? prev->so_q : prev->so_q0;
 177:         if (next == so)
 178:             break;
 179:         if (next == head)
 180:             return (0);
 181:         prev = next;
 182:     }
 183:     if (q == 0) {
 184:         prev->so_q0 = next->so_q0;
 185:         head->so_q0len--;
 186:     } else {
 187:         prev->so_q = next->so_q;
 188:         head->so_qlen--;
 189:     }
 190:     next->so_q0 = next->so_q = 0;
 191:     next->so_head = 0;
 192:     return (1);
 193: }
 194: 
 195: /*
 196:  * Socantsendmore indicates that no more data will be sent on the
 197:  * socket; it would normally be applied to a socket when the user
 198:  * informs the system that no more data is to be sent, by the protocol
 199:  * code (in case PRU_SHUTDOWN).  Socantrcvmore indicates that no more data
 200:  * will be received, and will normally be applied to the socket by a
 201:  * protocol when it detects that the peer will send no more data.
 202:  * Data queued for reading in the socket may yet be read.
 203:  */
 204: 
 205: socantsendmore(so)
 206:     struct socket *so;
 207: {
 208: 
 209:     so->so_state |= SS_CANTSENDMORE;
 210:     sowwakeup(so);
 211: }
 212: 
 213: socantrcvmore(so)
 214:     struct socket *so;
 215: {
 216: 
 217:     so->so_state |= SS_CANTRCVMORE;
 218:     sorwakeup(so);
 219: }
 220: 
 221: /*
 222:  * Socket select/wakeup routines.
 223:  */
 224: 
 225: /*
 226:  * Queue a process for a select on a socket buffer.
 227:  */
 228: sbselqueue(sb)
 229:     register struct sockbuf *sb;
 230: {
 231:     register struct proc *p;
 232:     extern int selwait;
 233: 
 234:     if ((p = sb->sb_sel) && (caddr_t)mfkd(&p->p_wchan) == (caddr_t)&selwait)
 235:         sb->sb_flags |= SB_COLL;
 236:     else
 237:         sb->sb_sel = u.u_procp;
 238: }
 239: 
 240: /*
 241:  * Wait for data to arrive at/drain from a socket buffer.
 242:  */
 243: sbwait(sb)
 244:     register struct sockbuf *sb;
 245: {
 246: 
 247:     sb->sb_flags |= SB_WAIT;
 248:     SLEEP((caddr_t)&sb->sb_cc, PZERO+1);
 249: }
 250: 
 251: /*
 252:  * Wakeup processes waiting on a socket buffer.
 253:  */
 254: sbwakeup(sb)
 255:     register struct sockbuf *sb;
 256: {
 257: 
 258:     if (sb->sb_sel) {
 259:         SELWAKEUP(sb->sb_sel, (long)(sb->sb_flags & SB_COLL));
 260:         sb->sb_sel = 0;
 261:         sb->sb_flags &= ~SB_COLL;
 262:     }
 263:     if (sb->sb_flags & SB_WAIT) {
 264:         sb->sb_flags &= ~SB_WAIT;
 265:         WAKEUP((caddr_t)&sb->sb_cc);
 266:     }
 267: }
 268: 
 269: /*
 270:  * Wakeup socket readers and writers.
 271:  * Do asynchronous notification via SIGIO
 272:  * if the socket has the SS_ASYNC flag set.
 273:  */
 274: sowakeup(so, sb)
 275:     register struct socket *so;
 276:     struct sockbuf *sb;
 277: {
 278:     register struct proc *p;
 279: 
 280:     sbwakeup(sb);
 281:     if (so->so_state & SS_ASYNC) {
 282:         if (so->so_pgrp < 0)
 283:             GSIGNAL(-so->so_pgrp, SIGIO);
 284:         else if (so->so_pgrp > 0 &&
 285:             (p = (struct proc *)NETPFIND(so->so_pgrp)) != 0)
 286:             NETPSIGNAL(p, SIGIO);
 287:     }
 288: }
 289: 
 290: /*
 291:  * Socket buffer (struct sockbuf) utility routines.
 292:  *
 293:  * Each socket contains two socket buffers: one for sending data and
 294:  * one for receiving data.  Each buffer contains a queue of mbufs,
 295:  * information about the number of mbufs and amount of data in the
 296:  * queue, and other fields allowing select() statements and notification
 297:  * on data availability to be implemented.
 298:  *
 299:  * Data stored in a socket buffer is maintained as a list of records.
 300:  * Each record is a list of mbufs chained together with the m_next
 301:  * field.  Records are chained together with the m_act field. The upper
 302:  * level routine soreceive() expects the following conventions to be
 303:  * observed when placing information in the receive buffer:
 304:  *
 305:  * 1. If the protocol requires each message be preceded by the sender's
 306:  *    name, then a record containing that name must be present before
 307:  *    any associated data (mbuf's must be of type MT_SONAME).
 308:  * 2. If the protocol supports the exchange of ``access rights'' (really
 309:  *    just additional data associated with the message), and there are
 310:  *    ``rights'' to be received, then a record containing this data
 311:  *    should be present (mbuf's must be of type MT_RIGHTS).
 312:  * 3. If a name or rights record exists, then it must be followed by
 313:  *    a data record, perhaps of zero length.
 314:  *
 315:  * Before using a new socket structure it is first necessary to reserve
 316:  * buffer space to the socket, by calling sbreserve().  This should commit
 317:  * some of the available buffer space in the system buffer pool for the
 318:  * socket (currently, it does nothing but enforce limits).  The space
 319:  * should be released by calling sbrelease() when the socket is destroyed.
 320:  */
 321: 
 322: soreserve(so, sndcc, rcvcc)
 323:     register struct socket *so;
 324:     int sndcc, rcvcc;
 325: {
 326: 
 327:     if (sbreserve(&so->so_snd, sndcc) == 0)
 328:         goto bad;
 329:     if (sbreserve(&so->so_rcv, rcvcc) == 0)
 330:         goto bad2;
 331:     return (0);
 332: bad2:
 333:     sbrelease(&so->so_snd);
 334: bad:
 335:     return (ENOBUFS);
 336: }
 337: 
 338: /*
 339:  * Allot mbufs to a sockbuf.
 340:  * Attempt to scale cc so that mbcnt doesn't become limiting
 341:  * if buffering efficiency is near the normal case.
 342:  */
 343: sbreserve(sb, cc)
 344:     struct sockbuf *sb;
 345: {
 346: 
 347: #ifdef FIX_43
 348:     if ((unsigned) cc > (unsigned)SB_MAX * CLBYTES / (2 * MSIZE + CLBYTES))
 349:         return (0);
 350: #else
 351:     if ((unsigned) cc > (unsigned)SB_MAX)
 352:         return (0);
 353: #endif
 354:     sb->sb_hiwat = cc;
 355:     sb->sb_mbmax = MIN(cc * 2, SB_MAX);
 356:     return (1);
 357: }
 358: 
 359: /*
 360:  * Free mbufs held by a socket, and reserved mbuf space.
 361:  */
 362: sbrelease(sb)
 363:     struct sockbuf *sb;
 364: {
 365: 
 366:     sbflush(sb);
 367:     sb->sb_hiwat = sb->sb_mbmax = 0;
 368: }
 369: 
 370: /*
 371:  * Routines to add and remove
 372:  * data from an mbuf queue.
 373:  *
 374:  * The routines sbappend() or sbappendrecord() are normally called to
 375:  * append new mbufs to a socket buffer, after checking that adequate
 376:  * space is available, comparing the function sbspace() with the amount
 377:  * of data to be added.  sbappendrecord() differs from sbappend() in
 378:  * that data supplied is treated as the beginning of a new record.
 379:  * To place a sender's address, optional access rights, and data in a
 380:  * socket receive buffer, sbappendaddr() should be used.  To place
 381:  * access rights and data in a socket receive buffer, sbappendrights()
 382:  * should be used.  In either case, the new data begins a new record.
 383:  * Note that unlike sbappend() and sbappendrecord(), these routines check
 384:  * for the caller that there will be enough space to store the data.
 385:  * Each fails if there is not enough space, or if it cannot find mbufs
 386:  * to store additional information in.
 387:  *
 388:  * Reliable protocols may use the socket send buffer to hold data
 389:  * awaiting acknowledgement.  Data is normally copied from a socket
 390:  * send buffer in a protocol with m_copy for output to a peer,
 391:  * and then removing the data from the socket buffer with sbdrop()
 392:  * or sbdroprecord() when the data is acknowledged by the peer.
 393:  */
 394: 
 395: /*
 396:  * Append mbuf chain m to the last record in the
 397:  * socket buffer sb.  The additional space associated
 398:  * the mbuf chain is recorded in sb.  Empty mbufs are
 399:  * discarded and mbufs are compacted where possible.
 400:  */
 401: sbappend(sb, m)
 402:     struct sockbuf *sb;
 403:     struct mbuf *m;
 404: {
 405:     register struct mbuf *n;
 406: 
 407:     if (m == 0)
 408:         return;
 409:     if (n = sb->sb_mb) {
 410:         while (n->m_act)
 411:             n = n->m_act;
 412:         while (n->m_next)
 413:             n = n->m_next;
 414:     }
 415:     sbcompress(sb, m, n);
 416: }
 417: 
 418: /*
 419:  * As above, except the mbuf chain
 420:  * begins a new record.
 421:  */
 422: sbappendrecord(sb, m0)
 423:     register struct sockbuf *sb;
 424:     register struct mbuf *m0;
 425: {
 426:     register struct mbuf *m;
 427: 
 428:     if (m0 == 0)
 429:         return;
 430:     if (m = sb->sb_mb)
 431:         while (m->m_act)
 432:             m = m->m_act;
 433:     /*
 434: 	 * Put the first mbuf on the queue.
 435: 	 * Note this permits zero length records.
 436: 	 */
 437:     sballoc(sb, m0);
 438:     if (m)
 439:         m->m_act = m0;
 440:     else
 441:         sb->sb_mb = m0;
 442:     m = m0->m_next;
 443:     m0->m_next = 0;
 444:     sbcompress(sb, m, m0);
 445: }
 446: 
 447: /*
 448:  * Append address and data, and optionally, rights
 449:  * to the receive queue of a socket.  Return 0 if
 450:  * no space in sockbuf or insufficient mbufs.
 451:  */
 452: sbappendaddr(sb, asa, m0, rights0)
 453:     register struct sockbuf *sb;
 454:     struct sockaddr *asa;
 455:     struct mbuf *rights0, *m0;
 456: {
 457:     register struct mbuf *m, *n;
 458:     int space = sizeof (*asa);
 459: 
 460:     for (m = m0; m; m = m->m_next)
 461:         space += m->m_len;
 462:     if (rights0)
 463:         space += rights0->m_len;
 464:     if (space > sbspace(sb))
 465:         return (0);
 466:     MGET(m, M_DONTWAIT, MT_SONAME);
 467:     if (m == 0)
 468:         return (0);
 469:     *mtod(m, struct sockaddr *) = *asa;
 470:     m->m_len = sizeof (*asa);
 471:     if (rights0 && rights0->m_len) {
 472:         m->m_next = m_copy(rights0, 0, rights0->m_len);
 473:         if (m->m_next == 0) {
 474:             m_freem(m);
 475:             return (0);
 476:         }
 477:         sballoc(sb, m->m_next);
 478:     }
 479:     sballoc(sb, m);
 480:     if (n = sb->sb_mb) {
 481:         while (n->m_act)
 482:             n = n->m_act;
 483:         n->m_act = m;
 484:     } else
 485:         sb->sb_mb = m;
 486:     if (m->m_next)
 487:         m = m->m_next;
 488:     if (m0)
 489:         sbcompress(sb, m0, m);
 490:     return (1);
 491: }
 492: 
 493: sbappendrights(sb, m0, rights)
 494:     struct sockbuf *sb;
 495:     struct mbuf *rights, *m0;
 496: {
 497:     register struct mbuf *m, *n;
 498:     int space = 0;
 499: 
 500:     if (rights == 0)
 501:         panic("sbappendrights");
 502:     for (m = m0; m; m = m->m_next)
 503:         space += m->m_len;
 504:     space += rights->m_len;
 505:     if (space > sbspace(sb))
 506:         return (0);
 507:     m = m_copy(rights, 0, rights->m_len);
 508:     if (m == 0)
 509:         return (0);
 510:     sballoc(sb, m);
 511:     if (n = sb->sb_mb) {
 512:         while (n->m_act)
 513:             n = n->m_act;
 514:         n->m_act = m;
 515:     } else
 516:         sb->sb_mb = m;
 517:     if (m0)
 518:         sbcompress(sb, m0, m);
 519:     return (1);
 520: }
 521: 
 522: /*
 523:  * Compress mbuf chain m into the socket
 524:  * buffer sb following mbuf n.  If n
 525:  * is null, the buffer is presumed empty.
 526:  */
 527: sbcompress(sb, m, n)
 528:     register struct sockbuf *sb;
 529:     register struct mbuf *m, *n;
 530: {
 531: 
 532:     while (m) {
 533:         if (m->m_len == 0) {
 534:             m = m_free(m);
 535:             continue;
 536:         }
 537:         if (n && n->m_off <= MMAXOFF && m->m_off <= MMAXOFF &&
 538:             (n->m_off + n->m_len + m->m_len) <= MMAXOFF &&
 539:             n->m_type == m->m_type) {
 540:             bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
 541:                 (unsigned)m->m_len);
 542:             n->m_len += m->m_len;
 543:             sb->sb_cc += m->m_len;
 544:             m = m_free(m);
 545:             continue;
 546:         }
 547:         sballoc(sb, m);
 548:         if (n)
 549:             n->m_next = m;
 550:         else
 551:             sb->sb_mb = m;
 552:         n = m;
 553:         m = m->m_next;
 554:         n->m_next = 0;
 555:     }
 556: }
 557: 
 558: /*
 559:  * Free all mbufs in a sockbuf.
 560:  * Check that all resources are reclaimed.
 561:  */
 562: sbflush(sb)
 563:     register struct sockbuf *sb;
 564: {
 565: 
 566:     if (sb->sb_flags & SB_LOCK)
 567:         panic("sbflush");
 568:     while (sb->sb_mbcnt)
 569:         sbdrop(sb, (int)sb->sb_cc);
 570:     if (sb->sb_cc || sb->sb_mbcnt || sb->sb_mb)
 571:         panic("sbflush 2");
 572: }
 573: 
 574: /*
 575:  * Drop data from (the front of) a sockbuf.
 576:  */
 577: sbdrop(sb, len)
 578:     register struct sockbuf *sb;
 579:     register int len;
 580: {
 581:     register struct mbuf *m, *mn;
 582:     struct mbuf *next;
 583: 
 584:     next = (m = sb->sb_mb) ? m->m_act : 0;
 585:     while (len > 0) {
 586:         if (m == 0) {
 587:             if (next == 0)
 588:                 panic("sbdrop");
 589:             m = next;
 590:             next = m->m_act;
 591:             continue;
 592:         }
 593:         if (m->m_len > len) {
 594:             m->m_len -= len;
 595:             m->m_off += len;
 596:             sb->sb_cc -= len;
 597:             break;
 598:         }
 599:         len -= m->m_len;
 600:         sbfree(sb, m);
 601:         MFREE(m, mn);
 602:         m = mn;
 603:     }
 604:     while (m && m->m_len == 0) {
 605:         sbfree(sb, m);
 606:         MFREE(m, mn);
 607:         m = mn;
 608:     }
 609:     if (m) {
 610:         sb->sb_mb = m;
 611:         m->m_act = next;
 612:     } else
 613:         sb->sb_mb = next;
 614: }
 615: 
 616: /*
 617:  * Drop a record off the front of a sockbuf
 618:  * and move the next record to the front.
 619:  */
 620: sbdroprecord(sb)
 621:     register struct sockbuf *sb;
 622: {
 623:     register struct mbuf *m, *mn;
 624: 
 625:     m = sb->sb_mb;
 626:     if (m) {
 627:         sb->sb_mb = m->m_act;
 628:         do {
 629:             sbfree(sb, m);
 630:             MFREE(m, mn);
 631:         } while (m = mn);
 632:     }
 633: }

Defined functions